"Are there any warning signs of when a bearing is about to fail?"
Think of a bearing failure as happening in four stages. During the first stage (or earliest detectable point using vibration analysis), the signals will appear in frequency bands around 250 to 350 kilohertz (KHz). In the second stage, a signal around 500 to 2,000 hertz (its natural frequency) will begin to ring.
At the onset of the third stage, the harmonics of the fundamental frequency will start to be very apparent. Defects in the races are now obvious and will be visible on vibration analysis of the noise signal. At this point, there will also be a significant temperature increase.
During the fourth stage, there will be very high vibration. The fundamental frequency and harmonics begin to decrease as the random ultrasonic noise is boosted. Temperatures will start to skyrocket as the bearing self-destructs.
So, the short answer is yes. There are definitely warning signs of a bearing about to fail. The real question is, "Do you know what to look for?" The most popular technologies today for bearing monitoring are vibration analysis, oil analysis, ultrasonics and thermography. We can use these tools to compare current states to historical data and accurately assess the remaining life of the bearing.
Vibration analysis and oil analysis are considered the best at predicting a failure but are not always the most cost-effective. Bearing manufacturers have long known of the relationship between bearing life and temperature. They even have formulas that work very well at calculating safe operating temperatures. These formulas and calculators show that once a bearing starts operating outside its ideal temperature range, its life will begin to degrade at an accelerated rate. (Keep in mind that for every 15 degrees C above 70 degrees C that the base oil operates, its life is more than halved.)
Knowing this, why is thermography not a more popular method for bearing life prediction? The monitoring of temperatures is not always considered reliable because of the sheer amount of variables that contribute to the heat generation. Ambient temperature, friction, speed variability, load and runtime all have an effect on the temperature that will be measured.
Friction is the variable you should be the most concerned with if trying to predict a failure, but how do you separate it from all the others? If you could account for all the variables accurately, the increases you would get in operating temperature could be a great indicator of an impending failure.
Perhaps the cheapest and easiest way to spot a bearing failure is to use a non-contact infrared thermometer. The caveat is that you must always account for the other variables as well.